Building Production-Ready AI Features - AI Factory 201

Building Production-Ready AI Features - AI Factory 201#

Who this is for#

  • Developers integrating Gen AI features into production applications

  • Data teams building pipelines for AI-driven Knowledge Bases

  • AI and platform teams configuring model serving and observability

  • Hybrid Manager users enabling Sovereign AI workloads at scale

---

General Goals#

By completing this path, you will:

  • Build multi-step AI Assistants and advanced Structures

  • Design Hybrid Knowledge Bases and multi-source RAG pipelines

  • Deploy and manage GPU-powered Model Serving

  • Implement observability and monitoring for AI-driven features

  • Learn patterns for production-grade governance and performance tuning

---

Modules by Focus Area#

---

1. Advanced Assistant & Structure Design#

Goals:

  • Implement Assistants with complex personas and memory

  • Create advanced multi-step Structures

  • Integrate Tools and external data flows

Estimated Time: ~30–45 min

Modules:

  1. Assistants Reference Manual — Design assistants with personas, memory, and tool usage patterns.

  2. Structures Reference Manual — Learn how to chain retrieval and actions into reliable flows.

  3. Create an Assistant in Gen AI Builder — Build a production‑ready assistant with tools and KBs.

  4. Create a Structure in Gen AI Builder — Implement multi‑step workflows with validations and branching.

  5. Create a Tool in Gen AI Builder — Wrap internal services and APIs as safe, governed tools.

---

2. Data Engineering & Hybrid Knowledge Bases#

Goals:

  • Design and manage Hybrid Knowledge Bases

  • Tune multi-source RAG pipelines

  • Implement metadata filtering and hybrid search

Estimated Time: ~30–45 min

Modules:

  1. Best practices for Hybrid Knowledge Bases — Patterns to balance freshness, coverage, and cost.

  2. Manage Knowledge Bases in Gen AI Builder — Operate KBs: re‑embedding, source lifecycle, and access control.

  3. Create a Retriever in Gen AI Builder — Tune search behavior, filters, and ranking at scale.

---

3. Model Serving with KServe#

Goals:

  • Deploy GPU-powered models

  • Tune runtime and resource settings

  • Understand the Model Serving lifecycle

Estimated Time: ~45–60 min

Modules:

  1. Model Serving Concepts

  2. How-To Configure a ClusterServingRuntime — Define runtimes to host families of models with shared settings.

  3. How-To Deploy an NVIDIA NIM Microservices available on build.nvidia.comcontainer with KServe — Use NVIDIA NIM images for high‑performance inference.

  4. How-To Update GPU Resources for an InferenceService — Right‑size GPU/CPU for performance and cost.

  5. How-To Verify InferenceServices and GPU usage — Validate endpoints and GPU usage in Hybrid Manager.

---

4. Observability & Monitoring#

Goals:

  • Implement observability for AI pipelines and Model Serving

  • Monitor performance and resource usage

  • Enable production readiness checks

Estimated Time: ~20–30 min

Modules:

  1. Observability for Model Serving — Metrics and logs for models and pipelines.

  2. How-To Monitor deployed models with KServe — Track health and performance of serving endpoints.

  3. Observability for Model Serving — Platform‑level dashboards and alerts.

---

Next steps#

After completing this 201 Path:

---

Contents